Simplify handling of mstatush

[Timmmm]

mstatush is not really handled correctly at the moment. The Mstatus type is:

bitfield Mstatus : xlenbits = {
  SD   : xlen - 1,

  // The MBE and SBE fields are in mstatus in RV64 and absent in RV32.
  // On RV32, they are in mstatush, which doesn't exist in RV64.  For now,
  // they are handled in an ad-hoc way.
  // MBE  : 37
  // SBE  : 36

  // The SXL and UXL fields don't exist on RV32, so they are modelled
  // via explicit getters and setters; see below.
  // SXL  : 35 .. 34,
  // UXL  : 33 .. 32,

  TSR  : 22,
...
  SIE  : 1,
}

register mstatus : Mstatus

There is an mstatush register:

bitfield Mstatush : bits(32) = {
  MBE  : 5,
  SBE  : 4
}
register mstatush : Mstatush

But it is never modified and we ignore writes to mstatush (because the model doesn't support big endian.

This is quite awkward. For other 64-bit registers, e.g. menvcfg the h version is just a view into the top 32 bits of the 64-bit value (search for Upper 32 bits of in the priv spec and you see a load of these). That is easy to handle because we just have a single bitfield Menvcfg : bits(64) and then read/write the relevant parts in the CSR access functions.

However mstatus is very annoyingly not quite like that because the SD bit is always at xlen - 1. It is in a different position on RV32 and RV64.

The best solution I can think of is this:

bitfield Mstatus : bits(64) = {
  SD_64   : 63, // SD bit for RV64.

  MBE  : 37
  SBE  : 36

  SXL  : 35 .. 34,
  UXL  : 33 .. 32,

  SD_32 : 31, // SD bit for RV32.

  TSR  : 22,
...
  SIE  : 1,
}

And then you have to do something like:

let sd = mstatus[if xlen == 32 then SD_32 else SD_64];

We can put that in a function.

I think that is better than the current situation, and it would remove the need to have getters/setters for MBE/SBE/SXL/UXL.

[Timmmm]

For reference, here are the latest fields for the registers.

It is only SD that moves, and SXL/UXL don't exist for RV32. The other fields are the same for both.